chore: cms docs

[kostasrim]

add cms docs

resolves dragonflydb/dragonfly#5445

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
documentation	Ready	Preview, Comment	Mar 20, 2026 8:04am

[vyavdoshenko]

1. CMS.INCRBY - incorrect behavior when key does not exist
   The docs claim that a new sketch is created with default dimensions if the key does not exist. This is incorrect - the command returns an error "CMS: key does not exist". Also missing: increment must be a strictly positive integer (> 0); passing 0 or a negative value returns an error.

2. CMS.INITBYPROB - inverted semantics of probability
   The docs describe probability as the probability that an estimate will fall within the error bounds. This is the opposite of the actual behavior - it is the failure probability (probability of exceeding the error bound). The source code confirms: depth = ceil(ln(1 / probability)), meaning a small value like 0.001 gives a deeper, more accurate sketch. The example value 0.999 produces depth = 1 (a single hash function), making the sketch practically useless.

3. CMS.MERGE - destination must be pre-initialized
   The docs do not mention that destination must already exist before calling the command. If it does not, the command returns "CMS: key does not exist". Should explicitly state: "The destination key must be pre-initialized via CMS.INITBYDIM or CMS.INITBYPROB."

4. CMS.MERGE - weights must be integers
   Weights are parsed as int64_t in the source code - fractional values are not supported. The docs describe them as "scaling factors" without specifying the type, which could mislead users into passing float values.

[kostasrim]

1. CMS.INCRBY - incorrect behavior when key does not exist
   The docs claim that a new sketch is created with default dimensions if the key does not exist. This is incorrect - the command returns an error "CMS: key does not exist". Also missing: increment must be a strictly positive integer (> 0); passing 0 or a negative value returns an error.
2. CMS.INITBYPROB - inverted semantics of probability
   The docs describe probability as the probability that an estimate will fall within the error bounds. This is the opposite of the actual behavior - it is the failure probability (probability of exceeding the error bound). The source code confirms: depth = ceil(ln(1 / probability)), meaning a small value like 0.001 gives a deeper, more accurate sketch. The example value 0.999 produces depth = 1 (a single hash function), making the sketch practically useless.
3. CMS.MERGE - destination must be pre-initialized
   The docs do not mention that destination must already exist before calling the command. If it does not, the command returns "CMS: key does not exist". Should explicitly state: "The destination key must be pre-initialized via CMS.INITBYDIM or CMS.INITBYPROB."
4. CMS.MERGE - weights must be integers
   Weights are parsed as int64_t in the source code - fractional values are not supported. The docs describe them as "scaling factors" without specifying the type, which could mislead users into passing float values.

Done take a look :)